arXiv:astro-ph/9702152v1 18 Feb 1997 · v ofG1 and thesurfacem assdensity...
Transcript of arXiv:astro-ph/9702152v1 18 Feb 1997 · v ofG1 and thesurfacem assdensity...
arX
iv:a
stro
-ph/
9702
152v
1 1
8 Fe
b 19
97
An estimate of H◦ from Keck spectroscopy of the gravitational lens system
0957 + 561
Emilio E. Falco & Irwin I. Shapiro
Harvard-Smithsonian Center for Astrophysics
60 Garden Street, Cambridge MA 02138
and
Leonidas A. Moustakas & Marc Davis
Astronomy Department, University of California, Berkeley, CA 94720
ABSTRACT
We present long-slit LRIS/Keck spectroscopic observations of the gravitational lens
system 0957 + 561. Averaged over all of our data, the rest-frame velocity dispersion σvof the central lens galaxy G1 is σv = 279 ± 12 km s−1. However, there appears to be
a significant decrease in σv as a function of distance from the center of G1 that is not
typical of brightest cluster galaxies. Within 0.′′2 of the center of G1, we find the average
σv = 316 ± 14 km s−1, whereas for positions > 0.′′2 from the center of G1, we find the
average σv = 266 ± 12 km s−1. A plausible explanation is that G1 contains a central
massive dark object of mass MMDO ≈ 4× 109h−1100M⊙ (h100 = H◦/100 km s−1 Mpc−1),
which contributes to the central velocity dispersion, and that the outer value of σv is
the appropriate measure of the depth of the potential well of G1. The determination
of a luminosity-weighted estimate of σv is essential for a determination of H◦ from
Q0957 + 561; our accurate measurements remove one of the chief uncertainties in the
assumed form of the mass distribution of the lens. Thus, with the recent apparent
reduction in the uncertainty in the measurement of the time delay for the images A
and B of Q 0957 + 561, ∆τBA = 417± 3 d (Kundic et al. 1996), we obtain an estimate
for the Hubble constant: H◦ = 62 ± 7 km s−1 Mpc−1. If for some reason the trend
of σv with slit position is spurious and we should use the dispersion averaged along
the slit, then the estimate of H◦ increases to 67 ± 8 km s−1 Mpc−1. These standard
errors do not, however, include any contribution from any errors in the assumed form
of the mass distribution of the lens. In particular, we used the mass model described
by Falco, Gorenstein & Shapiro (1991), as updated by Grogin & Narayan (1996a,
b). The reduced χ2 of model fits to the available position and magnification data
for this system is relatively high (∼ 4), indicating that the estimate of H◦ may have
a significant contribution from model errors. Further observations, discussed herein,
should allow such errors to be estimated reliably.
Subject headings: Gravitational Lenses — Dark Matter —
Quasars: individual (0957 + 561)
– 2 –
1. Introduction
Multiply-imaged quasars are excellent laboratories for the estimation of cosmological
parameters, particularly H◦. Lens systems showing time variability are essential, since
measurements of time delays for lensed image pairs are needed to scale from redshifts and angles
to physical length scales, as first discussed in two seminal papers by Refsdal (1964, 1966).
The gravitational lens system 0957 + 561 was the first to be discovered (Walsh, Carswell, &
Weymann 1979). It is observed to consist of two images A and B of a QSO at z = 1.41, and a
lens containing primarily a cluster of galaxies at z = 0.36 (Young et al. 1981; Angonin-Willaime,
Soucail & Vanderriest 1994, hereafter ASV94; Garrett, Walsh & Carswell 1992). This cluster is
thought to be responsible for the wide (∼ 6′′) separation of the images. The brightest member of
the cluster is a giant elliptical galaxy, G1, only ≈ 1′′ North of image B (Stockton 1980). Because it
is bright, and therefore presumably massive, and near B, G1 plays a critical role in the lensing of
Q 0957 + 561. There is also evidence that another cluster of galaxies (z ∼ 0.5) lies along the line
of sight to the QSO (ASV94); at the present level of accuracy of the data and of the lens models,
ignoring the fact that this cluster is at a different redshift has a negligible effect on our estimate
of H◦ (also, see Fischer et al. 1996).
Both QSO images have strong radio emission; VLBI brightness maps exhibit single jets whose
positions, structures and orientation angles have been accurately determined (Gorenstein et al.
1988b; Garrett et al. 1994; Campbell et al. 1994). The QSO images also radiate in X-rays and
exhibit there rather dramatic time variability (Chartas et al. 1995). The images of Q 0957 + 561
have been closely monitored for radio (Lehar et al. 1992; Haarsma et al. 1996) and optical (Schild
& Cholfin 1986; Vanderriest et al. 1989; Schild 1990; Schild & Thomson 1995; Kundic et al. 1995,
1996) variability since their discovery. Their light curves reveal variability with modest amplitudes
(< 20% over the previous decade), and there has been a continuing controversy regarding the
correct estimate from these variations of the time delay ∆τBA for A and B. Such estimates have
been either approximately 1.5 yr (Press, Rybicki & Hewitt 1992a, 1992b), or 1.1 yr (Schild 1990;
Schild & Thomson 1995; Pelt et al. 1996), with the most recent evidence now favoring the latter
value (Kundic et al. 1996; Haarsma et al. 1996).
Detailed modeling of 0957 + 561 has been extensively carried out in recent years (Falco,
Gorenstein & Shapiro 1991, hereafter FGS91; Grogin & Narayan 1996a, hereafter GN96a, see also
an erratum: Grogin & Narayan 1996b, hereafter GN96b). The modeling results point clearly to
two types of measurements required to convert an observed time delay into a value of H◦: the
central velocity dispersion σv of G1 and the surface mass density of the cluster (FGS91; GN96a).
These measurements are complementary and can be used to break a degeneracy in lens models
that describe the cluster and G1 (Gorenstein, Falco & Shapiro 1988a). Eliminating this degeneracy
can lead to a robust estimate of H◦. However, systematic errors may affect the conversion of σv to
a lensing mass, as discussed by Kochanek (1991) and in GN96a. For example, anisotropy in the
velocity distribution of stars in G1 affects the relationship between observed dispersion and actual
– 3 –
dispersion. Measurements of the velocity dispersion and of the light distribution of G1 with much
higher SNR than presently available would be required to study these effects.
We have concentrated on the measurement of σv, for which Rhee (1991) obtained a value of
303 ± 50 km s−1, with the 4.2−m William Herschel Telescope. The measurement is difficult due
to the proximity of image B to G1, and to the relatively high redshift of G1. We took advantage of
the large collecting area of the 10−m Keck I telescope and of the capabilities of the Low-Resolution
Imaging Spectrograph (LRIS; Oke et al. 1995), a combination that proved sufficient to obtain a
useful new result, as we report below. We used the same absorption line triplet, Mg I b, as Rhee
(1991), because it is relatively free of contamination from QSO and telluric lines (Stevenson 1994).
We describe our observations and our calibration of velocity dispersion measurements in §2.
In §3, we conclude with a discussion of our results and their significance regarding estimates of H◦.
2. Observations
We obtained spectra of G1, three template stars (bright K giants), and M81 with LRIS on
Keck I on 1995 March 28. We used the 600 g mm−1 grating (blazed at 7500 A) with a 1′′ × 180′′
slit (the same slit width as that of Rhee 1991) for all our observations. The spectral coverage
was 5098−7696 A for the stellar templates and M81, and 6149−8747 A for G1. The latter range
includes the Mg I b absorption triplet with rest λ ≈ 5175 A, which corresponds to λ ≈ 7038 A for
z ≈ 0.36. The detector was a Tektronix CCD with 20482 24−µm pixels, each subtending 0.′′213 on
the sky (Oke et al. 1995); the corresponding dispersion was 1.3 A/pixel. Table 1 shows a log of
the observations, listed in the order of their acquisition. The table includes the seeing for each
of our exposures, which we estimated as the mean of three measurements of the spatial FWHM
of unresolved objects (the A image and a nearby star), one each at the red and blue ends of the
spectrum, and one at intermediate wavelengths. Figure 1 is a diagram of the orientation of the
slit during our exposures. In addition to the object exposures, we obtained calibration arc spectra
with Hg-Ne-Ar lamps, as well as internal flats and biases, as usual for CCD calibrations. The slit
and spectrograph combination yielded a spectral resolution of 5.9± 0.2 A, estimated as the mean
FWHM of isolated lines in our calibration spectra.
We tested for measurable effects of flexure in exposures 1−4 by measuring the position of the
peaks and the FWHM of unresolved sky emission lines at 6365 A, 6950 A, 7370 A, and 7750 A.
We determined centers and FWHM for these lines at three different spatial positions along the
slit, one at the midpoint of the A−B line (which is nearly N−S), and one each ∼ 10′′ N and S of
this location. We found that the line centers shifted from exposure to exposure by < 1 A, and by
∼ 2 A from exposure 1 to exposure 4. The mean FWHM of the lines within each exposure was
again ∼ 5.9 A, identical to the width of the comparison lines obtained with very short exposures.
If the lines were broadened by 1 A flexure during each exposure, the resulting FWHM of sky lines
would be changed by too small a value for us to detect. A 2 A broadening would have resulted
– 4 –
in sky linewidths of order 6.2 A, marginally inconsistent with the measured linewidths. If our
data were contaminated by 1 A flexure, a 32 km s−1 width should be subtracted in quadrature
from the measured rest frame velocity dispersion; for the measured dispersion derived below such
contamination would make only a 2 km s−1 difference in our final answer.
The CCD frames were processed with the LRIS software package developed by Drew Phillips
(Lick Observatory), for bias subtraction and flat-fielding (only the long G1 exposures were
flat-fielded), taking into account the two-amplifier readout mode that we used with the CCD.
We reduced the spectra with the standard IRAF1 “longslit” package. We derived a wavelength
solution from the arc exposures (using task “identify” along CCD rows, and fitting Legendre
polynomials of up to 4th order), and determined the spatial distortion of the spectra from the
object exposures (using tasks “identify” and “reidentify” along CCD columns). We calculated
wavelength calibration and spatial distortion solutions with task “fitcoord” (using Chebyshev
polynomials of 4th and 3rd order along CCD lines and columns, respectively), and rectified the
spectra with task “transform.” Part of the processing with “transform” included rebinning the
spectra both linearly and logarithmically along the spectral direction (the latter for ultimate
analysis using the Fourier cross-correlation method, as we describe below).
For the G1 exposures, the slit was always centered on the B image of the QSO, in an attempt
to obtain simultaneously the best possible measurement of the spectrum of the QSO. In exposures
1−3, the slit was oriented N−S (PA = 180◦ for exposure 3), while in exposure 4, the slit was tilted
11◦ W of N−S, to obtain the best possible spectrum of image A. However, exposures 1−3 still
include a significant amount of light from A. We found the separation for QSO images A and B on
exposure 4 to agree well with the previous VLBI measurement (Gorenstein et al. 1988b) at the
∼ 0.′′01 level, thus confirming our choice of slit orientation at the ∼ 3◦ level.
We subtracted the sky background from exposures 1−4 with task “background.” The spatial
distortion of the spectra was significant; therefore we fit the sky only in two 15-pixel fitting strips,
one just below B and one just above A, and fit a slope to the background in the spatial direction
of the two strips. Because we are interested in the Mg I b triplet, the complicated structure of
OH emission in the neighborhood of 7000 A made it essential to subtract the sky accurately in
the ±10′′ vicinity of B. We determined the quality of our sky fits by checking the residuals as
a function of distance from B to its South, and from A to its North. Figures 2 and 3 show sky
spectra extracted from exposure 1 at ∼ 9′′ and ∼ 4′′ S of B, as well as the residuals after the same
sky subtraction procedure that we applied to our object spectra; these are typical of our results.
The magnitude of the residuals is consistent with our noise level, except where strong sky lines are
present; such regions are excluded from our velocity dispersion estimates. We found that, as can
be seen in the figures, the sky spectra are dominated by OH line emission. There is no significant
1IRAF (Image Reduction and Analysis Facility) is distributed by the National Optical Astronomy Observatories,
which are operated by the Association of Universities for Research in Astronomy, Inc., under contract with the
National Science Foundation.
– 5 –
telluric absorption in the region of interest. Outside of the OH bands, the sky-subtracted spectra
have noise per pixel that is approximately equally divided between sky noise and detector noise.
(The amplifier gain is approximately 1.6 e− DN−1 and the readout noise is approximately 8 e−.)
In the spatial region of chief interest, the sky subtraction is acceptable even throughout the OH
emission band. We tried subtracting a parabola instead of a slope, and narrowing or widening the
fitting strips, but found no significant change in the quality of the subtraction. For estimation of
the velocity dispersion described below, we worked in the narrowly defined band, 6912 − 7220 A,
to avoid the complications of the poorly subtracted OH bands outside this range.
Figure 4 shows spectra of B and A extracted from exposure 4, which was aligned with the
B−A line, and therefore yielded the QSO spectra of highest signal to noise ratio (SNR) in our set.
Table 2 shows a list of narrow absorption lines in our spectra of B and A. These lines, as well as a
broad Mg II emission feature, had been detected previously (Weymann et al. 1979; Wills & Wills
1980). Our measurements of absorption line centers and equivalent widths (with the spectral-line
measurement facilities in “splot” within IRAF) agree within the standard error (∼ 0.6 A RMS, or
∼ 26 km s−1 at 7000 A) with the previously published estimates. Our measurements of emission
line centers are more uncertain (∼ 5 A RMS) because the Mg II absorption lines in the wings of
the emission add significant uncertainty. The spectra also show possible blends of Fe II emission
from the QSO, at λobs ∼ 6300 and 7140 A.
Figures 5−8 show spectra that we extracted from exposures 1−4 from single rows in a strip of
the CCD that, projected on the sky, corresponds to separations between each row and the optical
center of G1 (Stockton 1980) from ∼ −0.′′4 to ∼ +0.′′5, negative (positive) signs indicating South
(North) of this center. We selected CCD rows starting from the row in which visual inspection
just revealed the presence of a Mg I b line, and ending with the row just before the one for which
visual inspection failed to detect the same line. We registered the different exposures with one
another by assuming that the CCD row with the largest peak brightness contained the center of
brightness of image B. A continuum was subtracted from each of these spectra by fitting cubic
splines of order 16, with 5 iterations to reject points with residuals higher than 2σ and lower
than 5σ from the fit, where σ is the RMS dispersion of each fit. The figures show the spectra
after the continuum subtraction. For each exposure, the total number of CCD rows that yielded
a useful spectrum (in the sense defined below) was 5, for a total of 20 useful spectra that we will
refer to hereafter as “G1 spectra.” Telluric O2 absorption is apparent in bands at 6870 A and at
7600 A, while telluric H2O absorption is present at 7240 A (Stevenson 1994). Fortunately, the
spectral features of interest for G1 do not fall on any of these bands and therefore no correction
was made for these absorption features. We checked that no telluric absorption is present in our
spectra in the region of interest, by examining spectra extracted from the slit on the North side
of image A; we are fortunate that this region is clean. In the vicinity of Mg I b, the G1 spectra
reveal additional absorption lines, which can be identified with Cr I 5208 A, Fe I 5227 A and
Ca I+Fe I 5270 A. These absorption lines affect our measurements at the ∼ 7 km s−1 level (see
our discussion of calibration of the velocity dispersion, below).
– 6 –
Note from Figures 5−8 that the G1 spectra extracted in the first row (∼ 0.′′4 from the optical
center of G1) are strongly contaminated by QSO emission that has leaked through the spline
continuum fit. We experimented with subtracting a scaled version of the spectrum of image A from
the individual G1 spectra, and found that such a procedure cleanly eliminates the contribution of
image B to the region centered at 6750 A, but that, as expected, the resulting G1 spectra have
measurably increased noise. Because the spectrum of the QSO is featureless in the narrow region
of interest, we decided against doing this subtraction in the final analysis.
We calculated the redshift of G1 and its line-of-sight velocity dispersion using the Fourier
cross-correlation method (Tonry & Davis 1979), as implemented in the IRAF task “fxcor.” Given
spectra of an object and of a template, fxcor yields estimates of the redshift of the object and
of the FWHM of absorption lines in the object spectrum. These quantities are estimated by
locating the peak in the cross-correlation and fitting a gaussian function to it, after background
subtraction. The redshift of the object spectrum is then inferred from the location of the peak
of the gaussian fit to the cross-correlation; the corresponding estimate of velocity dispersion is
related to the FWHM of the gaussian, as we describe below.
Our strategy with fxcor was to: (1) subtract continua from the spectra of the templates, in
the same fashion as for the G1 spectra; (2) restrict the region of the spectrum of G1 included in
the cross-correlation to ∼ 6900 − 7220 A and that of the templates to ∼ 5090 − 5330 A, to avoid
contributions from the QSO and poorly-subtracted telluric emission lines; (3) apodize the spectra
by 5% (at each end) to minimize spectral aliasing effects; (4) restrict the gaussian fits to the
central 21 pixels of the cross-correlation peak, to minimize the influence of the noise background
(varying this fitting width between 9 and 27 pixels affects our estimates by ≤ 5 km s−1; we
include this value in our error budget as a systematic contribution); (5) filter out the lowest
frequencies (wavenumbers ≤ 130 A−1; compare with Franx, Illingworth & Heckman 1989) present
in the Fourier-transformed template spectra, to remove any remaining low-frequency background.
Our template spectra and galaxy spectra have different resolutions as measured in km s−1
(see, e.g., van Dokkum & Franx 1996). The mismatch occurs because the resolution of the
spectrograph is nearly constant in A and the template is at rest, whereas G1 is redshifted to
z ≈ 0.36. Thus, the resolution in km s−1 of our template spectra (FWHM ∼ 340 km s−1 at
5200 A) is coarser than that of our G1 spectra (FWHM ∼ 250 km s−1 at ∼ 7072 A) by a factor
of ∼ 1.36. To allow us to estimate the effect of such a mismatch on our estimates of velocity
dispersions, M. Franx kindly provided spectra of template stars that D. Fisher (U.C. Santa Cruz)
acquired with the Kast Spectrograph on the Lick Observatory 3−m telescope on 1993 May 16.
These spectra covered the range from 4214 to 5620 A, with resolution ∼ 186 km s−1 at 5200
A, ∼ 1.3× the resolution of our G1 spectra. Thus, we use these spectra to avoid the resolution
mismatch. We selected spectra of three of the stars, HD126778, HD172401 and HR4435, where
the Mg I b line was as prominent as in our LRIS templates. We followed the procedure of van
Dokkum & Franx (1996), wherein before cross-correlation, the template spectra are rebinned and
smoothed to match the resolution of the G1 spectra in km s−1.
– 7 –
We obtained LRIS and Lick average templates by summing the corresponding spectra (both
smoothed and unsmoothed in the Lick case). We calibrated our calculations of velocity dispersions
by convolving the LRIS and Lick average template spectra with a series of 11 gaussians with σvincreasing in steps of 50 km s−1, from 0 to 500 km s−1. We then ran fxcor for each of these, with
the same parameters and wavelength sampling as those for G1. Figure 9 shows a calibration plot of
the FWHM of the output cross-correlation peak as a function of “input” velocity dispersion, both
in km s−1, for the LRIS and Lick templates. The velocity dispersion σv of G1 can be obtained
from such curves, by first blueshifting the galaxy spectra to z = 0 (see below for our estimate of
zG1), then running fxcor on these rest spectra, and finally by interpolating the calibration curves.
The plot shows calibrations for the averages of our LRIS templates and of our Lick templates,
both smoothed and unsmoothed. The figure shows that our procedure is indeed sensitive to the
mismatch in resolutions of the spectra of the templates and of G1. As can be seen in Figure 9, for
the average G1 spectrum and the average LRIS template (unsmoothed average Lick template),
the “output” FWHM of the cross-correlation peak is ∼ 813 km s−1(∼ 662 km s−1), and the
corresponding “input” σv is ∼ 238 km s−1(∼ 287 km s−1). For the smoothed average Lick
template, the “output” FWHM and input σv are ∼ 683 km s−1and ∼ 279 km s−1, respectively.
We adopt the curve for the smoothed average Lick template as the calibration of our velocity
dispersions.
We summed the individual rows with G1 spectra to maximize the SNR. Thus, we obtained a
mean G1 spectrum, as well as averages by exposure and by row, to estimate the RMS spread of
our measurements. We also repeated the velocity dispersion estimation procedure using in turn
each of the 3 stars in Table 1 as a template, with the G1 spectra in their role of object spectra.
For each template, we found the mean and the RMS spread of the FWHM estimate for the mean
G1 spectrum. For each of the four separate exposures, we added the data from the five useable
rows of CCD data. Table 3 lists the derived σv of the individual exposures relative to the mean
of the three template stars. We list the measured FWHM of the cross-correlation routine, and
the inferred σv for each measurement, as well as the R−value, indicative of the SNR for each
cross-correlation peak that we used (Tonry & Davis 1979). The largest source of uncertainty in
our velocity estimates is the variation from exposure to exposure. The mean σv averaged over the
four exposures is 279 ± 9 km s−1, where the error is the standard error of the mean of the four
independent observations. This error represents a fair estimate of the statistical uncertainty of our
velocity dispersion measurement.
Also listed in Table 3 are the results from the cross-correlation of the sum of all 20 rows of G1
data (shown in Figure 10; see also Figure 11) and each of the 3 template stars. Again, the results
are very consistent, and the standard deviation of the mean of σv for G1 is 2 km s−1. We tested
for the effects of the Cr I 5208 A, Fe I 5227 A and Ca I+Fe I 5270 A lines in the wavelength
regions that we used for cross-correlations. We found that restricting these regions even further, to
include only the Mg I b triplet, changed our mean velocity dispersion estimate to ∼ 272 km s−1.
Thus, the presence of these lines adds (in quadrature) a ∼ 7 km s−1 standard error to our error
– 8 –
budget. These are reasonable estimates of the systematic errors due to the possible mismatch of
the template spectrum to the intrinsic spectrum of G1. We sum all the entries in our error budget
in quadrature (see Table 4), arriving at a final estimate of σv = 279 ± 12 km s−1. However, this
standard error does not include any contribution from a possibly spatially-dependent variation in
the mass-to-light ratio in G1.
Finally, Table 3 lists the results from the cross-correlation with the average of each of the five
rows with useable spectra that we extracted from our exposures. We summed each row over all
four exposures to calculate the cross-correlations, to check whether our data yielded a detected
variation of the velocity dispersion of G1 as a function of the distance of the strip from the center
of brightness of G1, which we assumed to be at the Stockton (1980) coordinates as measured from
image B. Table 3 and Figure 12 show a seemingly very significant trend with position from the
center of G1. The χ2 fit for a null hypothesis, that all the dispersion measures are consistent with
a constant value, is ∼ 22.2 for 4 degrees of freedom and absent significant systematic error, can
be ruled out with high confidence. For rows 2 and 3, both within 0.′′2 from the center of G1, we
measure a mean σv = 316 ± 14 km s−1, whereas for rows 1, 4, and 5, all at more than 0.′′2 from
the center of G1, we measure σv = 266 ± 12 km s−1 (including all sources of error). The average
σv derived from the 5 rows is ∼ 286 km s−1, larger than σv averaged over exposures, but within
our standard errors.
With fxcor, we also measured the redshift of G1: zG1 = 0.356 ± 0.002. Our average template
spectrum is compared with that of the average of G1 in Figure 10, after convolution with a
gaussian with σ = 279 km s−1. As an additional reference, in Figure 11, we plot the spectrum of
the smoothed average Lick template, both before and after convolution with the gaussian.
As a partial check of our procedure, we determined the central velocity dispersion of M81,
σv = 180 ± 10 km s−1, which compares well with previous estimates (e.g., 177 ± 13 km s−1;
Keel 1989), but which does not suffer from the resolution mismatch that we had for G1. We also
utilized the program “four” kindly provided by M. Franx to obtain independent estimates of σv.
These estimates agreed with ours to one more figure than is significant.
3. Discussion and Conclusions
As first shown in Falco, Gorenstein & Shapiro (1985; see also FGS91 and GN96 a,b), a
degeneracy exists in lens models, between mass in the cluster and in G1. Any lens model can be
transformed by adding a sheet with uniform convergence κ < 1 to the cluster, and reducing the
mass of G1 by a factor (1− κ). Such a transformation maintains the observables invariant, except
for ∆τBA. Thus, by obtaining a value for the mass of G1, one is able to break the degeneracy, and
with an estimate for ∆τBA, to obtain an estimate of H◦. Because the mass of G1 can be estimated
from σ2v , our measurement breaks the degeneracy. However, the models of FGS91 and GN96a, b
assume that there is no significant additional variation in shear on the 6′′ scale of the separation
– 9 –
of the images; as noted below, this assumption may be somewhat in error.
As described by FGS91 and GN96a, and as updated by GN96b, the model-fitting now favors
the FGS91 mass model, which yields the following estimate of H◦ for the Q 0957 + 561 system:
H◦ = 98+12−11
(
σv330 km s−1
)2 (1.1 yr
∆τBA
)
km s−1 Mpc−1,
where the errors are 2σ bounds, as defined in GN96b, and do not include any contribution from
the uncertainty in the mass model. In the case σv = 330 km s−1, the galaxy would provide all
the convergence, while the cluster would provide none. Our average rest-frame velocity dispersion
estimate, σv = 279 ± 12 km s−1, is consistent with the value obtained by Rhee (1991) with a slit
of the same width as ours, but with significantly lower SNR. The most recent estimate of ∆τBA is
417 ± 3 d (Kundic et al. 1996). Using our average value of σv and that of Kundic et al. (1996)
for ∆τBA yields an estimate for the Hubble constant of H◦ = 67 ± 8 km s−1 Mpc−1, where the
error now includes the total observational uncertainty, but not the model uncertainty. This result
appears to be competitive in accuracy with those obtained from observations of objects at lower
redshifts.
The magnitude of the measured average decrease of σv with angular distance from the center
of G1 (both toward and away from image B) possibly suggests the existence of a MDO within
G1, but data of significantly higher angular resolution is needed (recall that zG1 = 0.36) before
one can estimate parameters for the MDO. The trend of σv does not seem quite consistent with
the gradual slope observed in brightest cluster galaxies (∆ log σv/∆ log r = −0.061 ± 0.004; Fisher
et al. 1995), and neither is the trend consistent with the expected signature of an MDO. If the
luminosity distribution of G1 were sufficiently “cuspy”, our spectra, taken in conditions of ∼ 0.′′7
seeing, might be contaminated by the increased dispersion this central mass would introduce. If
we take as a (crude) measure of the mass of the central MDO the virial mass corresponding to
(a) the velocity dispersion, σMDO, obtained by the quadrature subtraction of the average outer
dispersion from the average inner dispersion; and (b) the assumption that the size of the relevant
light distribution contributing to the inner dispersion is 0.′′2 (≈ 0.6h−1100 kpc at G1), we obtain
MMDO ≈ 4 × 109h−1100M⊙, because σMDO = 171 ± 20 km s−1. Note that the FGS91 model (as
updated in GN96b) also suggests the existence of an MDO within G1, but the inferred MMDO is
negligible compared to the mass required to explain the Q 0957 + 561 image splitting. Note also
that our MMDO estimate is comparable to those inferred for MDOs within nearby galaxies such
as M87 (∼ 2.4 × 109M⊙; Ford et al. 1994) and NGC3115 (∼ 109M⊙; Kormendy et al. 1996).
The estimate for MMDO may be considered an upper limit, because the light of G1 may well be
more concentrated than we assumed; the estimate might be improved with an HST image of the
galaxy (which recently became available from the HST archives). Thus, it may not be appropriate
for estimation of H◦ to use the average σv, if it is influenced appreciably by an MDO in the
nucleus. Assuming that rows 1, 4, and 5 yield the appropriate measure of the velocity dispersion,
we find that σv = 266 ± 12 km s−1 corresponds to the total mass of G1, and that H◦ = 62 ± 7
km s−1 Mpc−1.
– 10 –
An accurate measurement of H◦ is fundamental for progress in cosmology, and the subject
is alive with activity. Our estimate of H◦ is consistent with results emerging from a number of
separate estimates based on widely differing methods (see, e.g., proceedings of the May 1996
conference on the “Extragalactic Distance Scale” at STScI); it also agrees with the rather broad
limits set by recent measurements of time delays for the PG 1115+080 quadruple gravitational
lens system (Schechter et al. 1996); more recently Keeton & Kochanek (1996) have inferred from
this system a value of H◦ = 60± 17 km s−1 Mpc−1.
The cluster’s convergence and the line-of-sight velocity dispersion of G1 are measured
independently, and so should give self-consistent values, in the sense that 1−κ = (σv/330 km s−1)2.
The cluster convergence we infer for σv = 266 ± 12 km s−1 is κ = 0.35 ± 0.06; for σv = 279 ± 12
km s−1, we find κ = 0.28±0.06. These values are consistent with the value found by Kundic et al.
(1996), using the cluster mass distribution modeled by Fischer et al. (1996), κ = 0.24 ± 0.14(2σ).
Mapping the shear field of the cluster with HST by measuring the distortions it induces
on the background field of galaxies, could lead to a better estimate of its surface mass density
(see, e.g., Tyson, Valdes & Wenk 1990; Tyson & Fischer 1995; Schneider & Seitz 1995; Seitz &
Schneider 1995; Squires et al. 1996a, b; Seitz & Schneider 1996), and thus provide an additional
check of the cluster convergence we infer. Such a test is currently underway (Rhee et al. 1996).
To help remove remaining model uncertainties, other measurements of the Q 0957 + 561
system would also be useful. For example, our observation should be repeated on a night with
better seeing, and in addition observations should be made with an E−W orientation of the slit.
A high-quality image in the optical or near-IR would be useful to judge the “cuspiness” of G1 and
to assess whether it has a central MDO or even a black hole.
A more reliable measurement of the velocity dispersion of the cluster would also be useful to
refine estimates of its virial mass; we have acquired spectra of ∼ 25 additional galaxies in the field
of Q 0957 + 561, which we plan to combine with the redshift measurements of ASV94 to improve
the estimate of the cluster velocity dispersion.
Improved modeling must also be pursued for the Q 0957 + 561 system. As noted above and
emphasized by GN96a, b, the poor reduced χ2 of the models is a cause for concern (see also
Kochanek 1991). The VLBI plus optical data now yield, in effect, 11 observational constraints;
the models already include 5 parameters (3 for the G1 galaxy, plus 2 for the cluster), leaving
only 6 degrees of freedom. Introduction of additional parameters via a more complex model
will further reduce the number of degrees of freedom, so one must be very judicious in the
interpretation of these more complex models, because a large number of parameters could be used
to fit any constraint, but their values would not necessarily indicate a meaningful range of physical
properties.
Until we have obtained more data (and completed the reduction of others), as described
above, and examined our models of Q 0957 + 561 in greater depth, this tantalizing measurement of
the fundamental cosmological parameter H◦ must be considered only suggestive and not definitive.
– 11 –
Acknowledgments
The W. M. Keck Observatory is a scientific partnership between the University of California
and the California Institute of Technology, made possible by the generous gift of the W. M. Keck
Foundation. This work was supported in part by NSF grants AST92-21540 and AST93-03527.
We thank M. Franx, C. Kochanek, M. Geller and S. Kenyon for useful comments. LM would like
to thank SAO for travel support to the CfA. We also thank the referee, Tomislav Kundic, for
insightful comments that helped us clarify our results.
REFERENCES
Angonin-Willaime, M., Soucail, G. & Vanderriest, C. 1994 (ASV94), A&A 291, 411
Campbell, R. M. et al. 1994, AJ 426, 486
Chartas, G. et al. 1995, ApJ 445, 140
van Dokkum, P. G. & Franx, M. 1996, MNRAS in press
Falco, E. E., Gorenstein, M. V. & Shapiro, I. I. 1991 (FGS91), ApJ 372, 364
Falco, E. E., Gorenstein, M. V. & Shapiro, I. I. 1985, ApJ 372, 364
Fischer, P., Bernstein, G., Rhee, G. & Tyson, J. A. 1996, AJ in press (astro-ph/9608117)
Fisher, D., Illingworth, G. & Franx, M. 1995, ApJ 438, 539
Ford, H. et al. 1994, ApJ 435, L27
Franx, M., Illingworth, G. & Heckman, T. 1989, ApJ 344, 613
Garrett, M. A., Walsh, D., & Carswell, R. F. 1992, MNRAS 254, 27
Garrett, M. A. et al. 1994, MNRAS 270, 457
Gorenstein, M. V., Falco, E. E. & Shapiro, I. I. 1988a, ApJ 327, 693
Gorenstein, M. V. et al. 1988b, ApJ 334, 42
Grogin, N. A. & Narayan, R. 1996 (GN96a), ApJ 464, 92
Grogin, N. A. & Narayan, R. 1996 (GN96b), erratum, ApJ 473, 570
Haarsma, D. B., Hewitt, J. N., Lehar, J. & Burke, B. F. 1996, electronic preprint astro-ph/9607080
Keel, W. C. 1989, AJ 98, 195
Keeton, C. & Kochanek, C. 1996, electronic preprint astro-ph/9611216
– 12 –
Kochanek, C. 1991, ApJ 382, 58
Kundic, T. et al. 1995, ApJ 455, 5
Kundic, T. et al. 1996, electronic preprint astro-ph/9610162
Kormendy, J. et al. 1996, ApJ 459, L57
Lehar, J. et al. 1992, ApJ 384, 453
Oke, J. B. et al. 1995, PASP 107, 375
Pelt, J., Kayser, R., Refsdal, S., & Schramm, T. 1996, A&A 305, 97
Press, W.H., Rybicki, G.B. & Hewitt, J.N. 1992a, ApJ 385, 404
Press, W.H., Rybicki, G.B. & Hewitt, J.N. 1992b, ApJ 385, 416
Refsdal, S. 1964, MNRAS 128, 295
Refsdal, S. 1966, MNRAS 132, 101
Rhee, G. 1991, Nature 350, 211
Rhee, G. et al. 1996, in Proceedings of the 173rd Symposium of the IAU: Astrophysical
Applications of Gravitational Lensing, Melbourne, Australia 1995, 49
Schechter, P. et al. 1996, electronic preprint astro-ph/9611051
Schild, R. E. & Cholfin, B. 1986, ApJ 300, 209
Schild, R. E. 1990, AJ 100, 1771
Schild, R. E. & Thomson, D. J. 1995, AJ 109, 1970
Schneider, P. & Seitz, C. 1995, A&A 294, 411
Seitz, C. & Schneider, P. 1995, A&A 297, 287
Seitz, S. & Schneider, P. 1996, A&A 305, 383
Squires, G., Kaiser, N., Babul, A., Fahlman, G., Woods, D., Newmann, D. M. & Bohringer, H.
1996a, ApJ 461, 572
Squires, G., Kaiser, N., Fahlman, G., Babul, A. & Woods, D. 1996b, ApJ 469, 73
Stevenson, C. C. 1994, MNRAS 267, 904
Stockton, A. 1980, ApJ 242, L141
– 13 –
Tonry, J. & Davis, M. 1979, AJ 84, 1511
Tyson, J. A., Valdes, F. & Wenk, R. A. 1990, ApJ 349, L1
Tyson, J. A. & Fischer, P. 1995, ApJ 446, L55
Vanderriest, C. et al. 1989, A&A 215, 1
Walsh, D., Carswell, R. F. & Weymann, R. J. 1979, Nature 279, 381
Weymann, R. J. et al. 1979, ApJ 233, L43
Wills, B. J. & Wills, D. 1980, ApJ 238, 1
Young, P. J. et al. 1981, ApJ 241, 507
This preprint was prepared with the AAS LATEX macros v4.0.
– 14 –
4. FIGURE CAPTIONS
Fig. 1: The Q0957 + 561 field, with labels indicating the images A and B, and the galaxy G1.
The diameters of the circles are ∼ 0.′′9, corresponding to our worst seeing (Table 1). Vertical solid
lines represent the slit for exposures 1−3; dotted lines at an angle of ∼ 11◦ from vertical represent
the slit for exposure 4. Horizontal dashed lines show the positions of the CCD rows that we used
in our analysis, as they would appear if they were projected on the sky.
Fig. 2: Sky spectrum 9′′ S of B (heavy curve) and residuals (light curve) after the usual sky
subtraction (see text). The label indicates the location of the Mg I b triplet, redshifted to z ∼ 0.36.
Fig. 3: Same as Figure 2, except for 4′′ S of B.
Fig. 4: Spectra of A (light curve) and B (heavy curve). Straight sections of the spectra are the
locations of poorly subtracted telluric lines. Absorption and emission lines (Table 3) are labeled.
Fig. 5: Background-subtracted spectra extracted from the region of exposure 1 from 0.′′64 to 1.′′49
N of B (top to bottom). From bottom to top, the ordinates of four successive spectra are shifted
down by 0.5, 1.0, 1.5 and 2.0 counts normalized by the continuum, relative to the northernmost
spectrum.
Fig. 6: Same as Figure 5, except for exposure 2.
Fig. 7: Same as Figure 5, except for exposure 3.
Fig. 8: Same as Figure 5, except for exposure 4.
Fig. 9: Calibration of the “output” FWHM of the fxcor gaussian fit to the cross-correlation peak
as a function of “input” dispersion σv in km s−1. The solid (dotted) curve is the calibration for
the smoothed average Lick (average LRIS) template. For comparison, the dashed curve shows the
calibration for the average Lick template without smoothing. As expected, the use of the dotted
curve taken with LRIS only data would lead to smaller estimates of the dispersion of G1. The
straight lines indicate the conversions of FWHM to σv for G1, for the average of our 20 useful
exposures (see the text). For an “input” of σv = 0 km s−1, each FWHM is approximately equal to
the FWHM of the cross-correlation peak of each template with itself, given the spectral resolution
in each case.
Fig. 10: Spectra of G1 (solid curve) and of the average smoothed Lick template star (dashed
curve). Each spectrum was normalized by its continuum for this comparison. The template was
convolved with a gaussian with σ = 279 km s−1, and G1 was blueshifted from z = 0.356 to rest.
Fig. 11: Spectra of the average Lick template star. Both spectra were normalized by their continua
for this comparison. The solid curve is the original smoothed average template; the dashed curve
shows the result of a convolution of the original smoothed curve with a gaussian with σ = 279
km s−1.
– 15 –
Fig. 12: Variation of the velocity dispersion as a function of the distance to the center of brightness
of G1. Each point represents one row extracted from our spectra, as explained in the text. Rows
North (South) of the center of brightness of G1 are represented by squares (triangles). The mean
of the values is indicated by the dashed line. Error bars indicate the standard error in the mean
for each row, which we estimated from our 20 rows with useful exposures.
– 16 –
5. TABLE CAPTIONS
Table 1: Observations log with LRIS/Keck I, 1995 March 28.
Table 2: Absorption and emission lines in the spectra of Q 0957 + 561 A, B.
Table 3: Cross-correlation results.
Table 4: Error budget for σv estimates.
– 17 –
Table 1: Observations log
Exposure Object Central λ Exposure UT airmass PA seeing FWHM
A sec ◦ E of N ′′
1 G1 7000 1500 8:02:56 1.24 0 0.90
2 G1 7000 1500 8:33:53 1.25 0 0.80
3 G1 7000 1500 10:46:56 1.49 180 0.75
4 G1 7000 1500 11:17:36 1.61 169 0.74
5 HD136711 5000 4 12:11:53 1.04 0 0.54
6 HD132737 5000 1 12:00:00 1.03 0 0.49
7 M81 5000 30 5:36:59 1.67 0 0.701
8 AGK2+14 783 5000 2 5:55:15 1.02 0 0.72
1 We assumed the seeing FWHM for M81 was comparable to that for AGK2+14 783, because there
were no unresolved objects in the spectra that we could use to obtain an estimate.
Table 2: QSO absorption and emission lines
A B
Ion λrest λobs Wλ λobs Wλ
A A A A A
Fe II 2599.4 6216.4 3.3 6215.9 3.5
Mg II 2797.5 6684.7 5.3 6684.0 5.3
Mg II 2802.5 6701.8 4.6 6701.3 4.6
Mg II 2852.1 6819.8 0.6 6819.8 0.8
Mg II1 2797.5 6757.3 −69.2 6758.2 −65.
1 The only strong QSO emission line is that of Mg II 2797.5 A. 2 Negative equivalent widths
correspond to emission lines.
– 18 –
Table 3: Cross-correlation results
Spectrum Template FWHM σv R−value
km s−1 km s−1
Exp. 1 average 696. 285. 13.
Exp. 2 average 723. 300. 12.
Exp. 3 average 661. 267. 15.
Exp. 4 average 660. 266. 11.
sum 1− 4 HD126778 672. 272. 12.
sum 1− 4 HR4435 679. 277. 12.
sum 1− 4 HD172401 676. 275. 12.
row 11 average 680. 277. 11.
row 2 average 785. 332. 11.
row 3 average 723. 300. 12.
row 4 average 649. 260. 15.
row 5 average 653. 262. 14.
1 Rows are separated by ∼ 0.′′213, starting ∼ 0.′′640 North and ending ∼ 1.′′49, also North of QSO
image B (see Fig. 1).
Table 4: Error budget for σv estimates.
error source standard error
km s−1
choice of template star 9
cross-correlation fitting width 5
choice of absorption lines 7
– 19 –
– 20 –
– 21 –
– 22 –
– 23 –
– 24 –
– 25 –
– 26 –
– 27 –
– 28 –
– 29 –